import pandas as pd 
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt
from datetime import date
import warnings

warnings.filterwarnings('ignore')

# data wrangling for the original data source

df = pd.read_csv("online_retail_listing.csv", delimiter = ';', encoding= 'unicode_escape')
df.dropna(how='any', inplace=True)
df.rename(columns= {'Price':'Unit Price', 'Country,,,,':'Country', 'InvoiceDate':'Order Date'}, inplace=True)

df['Unit Price']=df['Unit Price'].str.replace(',','.')
df['Unit Price']=df['Unit Price'].astype('float64')
df['Quantity']=df['Quantity'].astype('int64')
df['Order Date']=df['Order Date'].astype('datetime64')

df['Today']= '2012-02-13 12:00:00'
df['Today']= df['Today'].astype('datetime64')
df['Monetary Value'] = df['Unit Price']*df['Unit Price']

df.drop_duplicates(inplace=True)
df.columns

Index(['Invoice', 'StockCode', 'Description', 'Quantity', 'Order Date',
       'Unit Price', 'Customer ID', 'Country', 'Today', 'Monetary Value'],
      dtype='object')

df['Today']= date.today() df['Today']= df['Today'].astype('datetime64') # to calculate real time interval from last order date to today

# integrate a dataset ready for calculation 

rfm=df[['Customer ID']]
rfm['First Order Date'] = df.groupby(['Customer ID'])['Order Date'].transform('min')
rfm['Last Order Date'] = df.groupby(['Customer ID'])['Order Date'].transform('max')

rfm['Interval'] = df['Today'] - rfm['Last Order Date']
rfm['Interval'] = rfm['Interval'].astype('timedelta64[D]')
rfm['Interval'] = rfm['Interval'].astype('int')

rfm['Lifetime'] = abs(rfm['First Order Date'] - rfm['Last Order Date'])
rfm['Lifetime'] = rfm['Lifetime'].astype('timedelta64[D]')
rfm['Lifetime'] = rfm['Lifetime'].astype('int')

rfm['Frequency'] = df.groupby(['Customer ID'])['Order Date'].transform('count')
rfm['Total Spent'] = df.groupby(['Customer ID'])['Monetary Value'].transform('sum').round(2)

rfm.drop_duplicates(inplace=True)
rfm.reset_index(inplace=True)
rfm.drop(['index'], axis=1, inplace=True)
rfm.sample(5)

	Customer ID	First Order Date	Last Order Date	Interval	Lifetime	Frequency	Total Spent
2035	14651.0	2010-03-18 11:39:00	2011-12-05 16:29:00	69	627	283	9583.83
689	17603.0	2009-10-12 09:44:00	2011-10-20 19:19:00	115	738	115	602030.72
5656	17832.0	2011-10-21 11:43:00	2011-10-21 11:43:00	115	0	60	179.03
4887	15952.0	2011-02-10 12:47:00	2011-06-11 11:18:00	247	120	84	1497.36
416	16104.0	2009-06-12 10:41:00	2011-09-10 10:23:00	156	819	241	5546.57

# rfm['Interval'] = np.array([date.today()], dtype='datetime64') - rfm['Last Order Date'] cannot add indices of unequal length

rfm["FrequencyScore"] = pd.cut(rfm["Frequency"], 
                                       bins=[-1, 
                                             np.percentile(rfm["Frequency"], 25), 
                                             np.percentile(rfm["Frequency"], 50), 
                                             np.percentile(rfm["Frequency"], 75), 
                                             rfm["Frequency"].max()], 
                                       labels=[1, 2, 3, 4]).astype("int")

rfm["MonetaryScore"] = pd.cut(rfm["Total Spent"], 
                                       bins=[-1, 
                                             np.percentile(rfm["Total Spent"], 25), 
                                             np.percentile(rfm["Total Spent"], 50), 
                                             np.percentile(rfm["Total Spent"], 75),
                                             rfm["Total Spent"].max()], 
                                       labels=[1, 2, 3, 4]).astype("int")

rfm["RecencyScore"] = pd.cut(rfm["Interval"], 
                                       bins=[-1, 
                                             np.percentile(rfm["Interval"], 25), 
                                             np.percentile(rfm["Interval"], 50), 
                                             np.percentile(rfm["Interval"], 75),
                                             rfm["Interval"].max()], 
                                       labels=[4, 3, 2, 1]).astype("int")

# calculate RFM scores 

rfm['RFM Score'] = rfm["FrequencyScore"] + rfm["MonetaryScore"] + rfm["RecencyScore"]

# segmentize customer group based on RFM scores

rfm["Loyalty"] = pd.cut(rfm["RFM Score"], 
                                       bins=[-1, 
                                             np.percentile(rfm["RFM Score"], 33), 
                                             np.percentile(rfm["RFM Score"], 66), 
                                             rfm["RFM Score"].max()], 
                                       labels=['Silver', 'Gold', 'Platinum']).astype("str")

rfm

	Customer ID	First Order Date	Last Order Date	Interval	Lifetime	Frequency	Total Spent	FrequencyScore	MonetaryScore	RecencyScore	RFM Score	Loyalty
0	13085.0	2009-01-12 07:45:00	2011-05-07 12:11:00	281	845	88	690628.40	3	4	2	9	Gold
1	13078.0	2009-01-12 09:06:00	2011-12-10 14:48:00	64	1062	821	19615.11	4	4	4	12	Platinum
2	15362.0	2009-01-12 09:08:00	2010-09-17 10:37:00	514	613	40	840.92	2	2	1	5	Silver
3	18102.0	2009-01-12 09:24:00	2011-11-28 12:55:00	76	1050	1027	22629772.57	4	4	4	12	Platinum
4	12682.0	2009-01-12 09:28:00	2011-12-08 15:22:00	66	1060	962	28539.73	4	4	4	12	Platinum
...	...	...	...	...	...	...	...	...	...	...	...	...
5918	12587.0	2011-02-12 12:47:00	2011-02-12 12:47:00	365	0	4	1612.98	1	3	2	6	Silver
5919	13560.0	2011-02-12 13:11:00	2011-02-12 13:11:00	365	0	32	358.05	2	2	2	6	Silver
5920	15539.0	2011-02-12 13:36:00	2011-02-12 13:36:00	365	0	41	101.37	2	1	2	5	Silver
5921	13017.0	2011-02-12 15:53:00	2011-02-12 15:53:00	365	0	1	18.06	1	1	2	4	Silver
5922	13153.0	2011-04-12 12:14:00	2011-04-12 12:14:00	306	0	15	42.29	1	1	2	4	Silver

5923 rows × 12 columns

# discover if there is a correlation between lifetime and customers' RFM scores

fig, ax = plt.subplots(figsize=(5, 5), constrained_layout=True)
ax.axis('on')
ax.bar(rfm['RFM Score'], rfm.groupby(['RFM Score'])['Lifetime'].transform('mean'))
ax.set_title('Average Lifetime for Each RFM Segments',
             fontdict={'fontsize': '17', 'fontweight' : '3'})
plt.show()

# visualization of current customer base composition

fig, ax = plt.subplots(1, 1, figsize=(5, 5))

sns.countplot(x="Loyalty", data=rfm, color='steelblue')
ax.bar_label(ax.containers[0])

[Text(0, 0, '1815'), Text(0, 0, '1697'), Text(0, 0, '2411')]

# Monetary vs Recency (combined)
fig, ax = plt.subplots(1, 1, figsize=(5, 5))
sns.scatterplot(x='Interval', y='Total Spent', data=rfm, hue='Loyalty', size='Total Spent', palette=["gold", "green", "red"])
ax.set_xlim(1, 800)
ax.set_ylim(1, 4000)

# Frequency vs Recency (combined)
fig, ax = plt.subplots(1, 1, figsize=(5, 5))
sns.scatterplot(x='Interval', y='Frequency', data=rfm, hue='Loyalty', size='Total Spent', palette=["gold", "green", "red"])
ax.set_xlim(1, 800)
ax.set_ylim(1, 2000)

# Monetary vs Frequency (combined)
fig, ax = plt.subplots(1, 1, figsize=(5, 5))
sns.scatterplot(x='Total Spent', y='Frequency', data=rfm, hue='Loyalty', size='Total Spent', palette=["gold", "green", "red"])
ax.set_xlim(1, 4000)
ax.set_ylim(1, 1000)

(1.0, 1000.0)

import plotly.express as px

# Monetary vs Recency (combined)
fig1 = px.scatter(rfm, x='Interval', y='Total Spent', color='Loyalty')
fig1.update_layout(yaxis_range=[1,4000], xaxis_range=[1,800])

# Frequency vs Recency (combined)
fig2 = px.scatter(rfm, x='Interval', y='Frequency', color='Loyalty')
fig2.update_layout(yaxis_range=[1,800], xaxis_range=[1,800])

# Monetary vs Frequency (combined)
fig3 = px.scatter(rfm, x='Total Spent', y='Frequency', color='Loyalty')
fig3.update_layout(yaxis_range=[1,500], xaxis_range=[1,4000])